6
$\begingroup$

I've seen a lot of talk on the SE metas (metae? metapodi?) about AI answers, but I haven't seen any methodology for detecting it. I've run a few answers (that looked suspicious: really well-written from new members) through an AI checker.

What is the formal approach for us (specifically as users, I don't have an interest in some secret SE algorithm) in detecting AI?

$\endgroup$
2
  • 6
    $\begingroup$ Are you aware of the false positive error rate of such screens? I have not checked, but a post on main meta (admittedly old but still official) debates the usefulness of detectors and concludes their high error rate makes them unreliable. OTOH AI generation features tend so far not to be hard to discern and user flags and comments are welcome. $\endgroup$ Commented Apr 25, 2024 at 18:58
  • 2
    $\begingroup$ Also How can we determine whether an answer used ChatGPT? $\endgroup$ Commented Apr 26, 2024 at 3:11

1 Answer 1

Reset to default
8
$\begingroup$

I would encourage users to evaluate the quality of the answer. Currently, AI-generated answers are mostly not that good, so down-vote low-quality answers. If you are fairly certain an answer is AI-generated, flag it with "in need of moderator intervention", adding you suspect generative AI use.

Depending on the tools moderators have available and how much of a signal they give, we might delete the flagged answer. If we don't delete it, at least the score indicates it is not the best answer.

In cases where you have a better answer, I encourage you to post it. With enough eyes on the AI-suspected answer and on your answer, yours might float to the top.

$\endgroup$
2
  • $\begingroup$ Ai is not bad as such, and increasing in quality to a degree that many human answers are of less perceived quality. I would suggest a new item in the "flag" menu, something like "suspected AI", because only moderators have an overview of all the actions. Honest AIs should address themselves as such, just as in the Asimov fiction en.wikipedia.org/wiki/R._Daneel_Olivaw . $\endgroup$ Commented Oct 27, 2024 at 20:37
  • $\begingroup$ @GyroGearloose Moderators aren’t experts in detecting AI but are pretty good at detecting bad answers. And customer-facing AI are still weak when it comes to chemistry. Part of the weakness might be the low quality of chemistry-related content on the internet. $\endgroup$ Commented Oct 28, 2024 at 2:09

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.